20 Disruptions, 20 Lessons
Ten general-purpose technologies and eleven occupation-level disruptions. Each case study follows the same arc: prediction, reality, mechanism, and what it means for AI.
What is a cross-case structured metric?
A cross-case structured metric is a single dimension (like “time to 50% task displacement”) populated consistently across multiple historical disruptions so cases can be compared on the same axis. Every case below is enumerated against the same five metrics where data feasibility allows: (a) Time to 50% Task Displacement, (b) Institutional Response Lag, (c) Reskilling Adjacency, (d) Geographic / Demographic Concentration, and (e) Peak Annual Displacement Rate. We use them because narrative-only comparison invites cherry-picking; structured metrics force every case to answer the same questions or declare why it can’t.
7 high-feasibility cases carry full metric values with derivation method and uncertainty band. 8 medium-feasibility cases carry actual values where data permits, paired with per-cell confidence flags (H / M / L). When a cell’s flag is U (uncertain) the value is suppressed rather than shown with false precision. 5 low-feasibility cases are treated narratively. Per BR-21, every derived value ships with derivation method and uncertainty band, not a clean number. The full feasibility matrix lives on findings.html.
⚠ Structural-bias warning
Reinstatement effects (the share of new tasks that absorb displaced workers) have weakened as headwinds over the last four decades. Historical augmentation patterns (e.g. spreadsheets → accountants) are calibrated against a stronger reinstatement rate than current AI deployment is likely to encounter. Treat historic Type 2 outcomes as upper bounds, not point predictions, for AI cases.
1. Macro Disruptions Ten general-purpose technologies that reshaped the labour market
Each GPT follows a 4-phase arc: market entry → mass adoption → productivity gains → full restructuring. The arc is compressing over time.
2. Micro Disruptions Eleven occupations, eleven different outcomes from the same forces
Each case study includes a disruption Type badge linking to the taxonomy on the overview page. Frey & Osborne (2013) estimated 47% of US jobs are susceptible to automation in principle; the OECD estimated 14% fully replaceable. The difference is methodological. But both agree: the inverse relationship between income/education and exposure is the dominant pattern. ISCO codes link to the AI Exposure Map.
3. Timescale Comparison How long disruption actually takes
Sorted by total span. The fastest completed disruption took 15 years; the slowest, over 90. Median: 30–40 years.